james-revisoai

12 ms·

91.
by james-revisoai 3y ago
I wrote more about this in another comment ("Postmortem"), as I developed a similar idea from Mid 2022 to early 2023, but made different decisions. OpenAI had a serious focus on this that did see adoption; as early as Spring 2021
92.
by james-revisoai 3y ago
Beware that most 2020-2021 OpenAI startups haven't survived to today, especially since ChatGPT allowed free use of most B2C use cases (article writing, paraphrasing, etc). I posted a brief postmortem of my experience from 2021 to 2023
93.
by james-revisoai 3y ago
Here's a post-mortem and the true history of this tooling approach and being a first mover, or not... Pre 2021: Useful semantic search widely deployed, used for recommendations for sales. OpenAI has an answers endpoint specifically for
94.
by james-revisoai 3y ago
1. Yes, very. Loading indicator could be a little more clearer on time estimate, but great/outstanding overall. 2. No, it seems too direct and not creative. I wonder if the original non-instruct models (curie etc) would be better for t
95.
by james-revisoai 3y ago
Specifically, we used semantic similarity for shuffled answers (so they are close and not too easy), completely shuffled answers, and AI-generated distractors. We are now going to choose the most popular default (based on retention during A
96.
by james-revisoai 3y ago
Really good points on the angle there, thank you. And important point that it shouldn't necessarily come from you! (that's what's worked better for us everywhere but ProductHunt, reflecting on it)
97.
by james-revisoai 3y ago
Thank you, this bug has been fixed. We recently changed from shuffled distractors to Ai-generated distractors, but as you identified, such a bug could occur.
98.
Ask HN: How can I improve how I Show HN?
6 points by james-revisoai 3y ago|8 comments
99.
by james-revisoai 3y ago
Can you expand on some good guard rail metrics you've used/seen please?
100.
by james-revisoai 3y ago
A flipside effect to this is advertising and PPC which has trended up so high overtime, especially in education. Strategies viable in 2019 are largely useless to bootstrapped businesses in 2023: the costs of ads even to acquire customers wi
101.
by james-revisoai 3y ago
Yes, that's correct. Then OpenAI bills you around the end of the month so you have some leeway of credit. You can use local embeddings, such as sentence transformer models, too, to save on embeddings (and maintain embeddings you can mo
102.
by james-revisoai 3y ago
It's also the norm at small German universities and, till recently, many larger ones. It has not harmed German students and I believe puts Germany in a better position right now.
103.
by james-revisoai 3y ago
It's an unusual 2 year or so window where this has been fairly possible, and still not realised or accepted by instructors. It's just too outside the current paradigm. It has coincided with students not having to do exams during C
104.
by james-revisoai 3y ago
Thank you. That's a good point, we should experiment soon. Originally, we had free text input, but too many students quit with that difficulty. Maybe a mix or new concept can fit well here. Edit: Re-reading your comment, perhaps it&#x2
105.
Show HN: Learn beautifully, with content-specific AI lessons(studywand.com)
5 points by james-revisoai 3y ago|2 comments
106.
by james-revisoai 3y ago
have you been on either side?
107.
by james-revisoai 3y ago
Request a GPU allocation, and use it, request again, etc. Unfortunately it has been like this since around 2020/2021. It's sensible on AWS' side I guess. If you really need them, you won't get credits but there are GPU-s
108.
by james-revisoai 3y ago
It's mentioned in the article, but LoRA or RAG will enable this. Phind is getting awfully close to this point already really. Integrating new knowledge isn't a bottleneck like we know from expert systems, I think it just hasn&#x27
109.
by james-revisoai 3y ago
You're right and this is why they didn't heavily use BERT(in the full sense), arguably the game-changing NLP model of the 10s. They couldn't justify bringing the cost per search up.
110.
by james-revisoai 3y ago
ChatGPT is expensive, but take Chegg for example. They hire thousands of people to support a system based on human-produced content delivered in the right time threshold to be worth paying for. Now ChatGPT is a lower-risk way (less likely t
111.
by james-revisoai 3y ago
Langchain --> query_with_sources function (semantic search context) If you finetune, you'll get mostly the style and sometimes worse factual performance. Source through semantic search context/RAG is a better approach. Source:
112.
by james-revisoai 3y ago
Thing is for things to take off, what is often key is retention. And retention stats relative to the normal/non-AI way of doing it (e.g. learning a wikipedia page) with these models for most use cases have looked like: GPT-2: 2% GPT-3:
113.
by james-revisoai 3y ago
This pretty heavily assumes that you can't learn programming by paper or on a shared school computer (quite widely available in the UK/US for a couple of decades now), which itself reveals a bias in your mindset here, tbh.
114.
by james-revisoai 3y ago
A great deal of advice here is going to push you in the direction of VC funded startups and the Lean Startup style approach. You might want to go that way, but as you said "build", not "sell", you may value the approach
115.
by james-revisoai 3y ago
The trend is also reversing in some segments Apple owned which masked that there may be quite a trend occurring for home purchases, but the data doesn't show it because the extra purchases and hidden by losses. For example University
116.
by james-revisoai 3y ago
The ScienceQA paper from amazon supports this. In a task where images and text were used for prediction tasks (and often the images showed physical things like pulling a door while the text was a question about physics etc), even the items
117.
by james-revisoai 3y ago
I built a 3D visualiser that overlays semantic topics and sentences of documents with different "embedding types and orders"[1] using Natural Language Processing. It colours each document with the same colour. You can see how two
118.
by james-revisoai 3y ago
To be fair, documents context with things like semantic search could be the upcoming bottleneck for a lot of industry applications. Even though a hundred+ QA products doing this (Chat with your files vibes etc) haven't had much success
119.
by james-revisoai 3y ago
It feels like this but every time we remove such an issue (and you're right, it's usually I/O or OCR/ML inference), we just get another. Whether it's the cloud blocking new GPU allowance increases for a week (I thin
120.
by james-revisoai 3y ago
Fair enough. I suppose in the current situation, we have too much demand and scaling isn't so much a business priority relative to other ongoing events, so it's kind of the startupy view of tech I'm looking for. I think you a

More ›